---
license: other
license_name: amuse-research-license
license_link: https://github.com/kiranchhatre/amuse/blob/main/LICENSE
extra_gated_heading: Request access to the AMUSE research artifacts
extra_gated_description: Access is manually reviewed. For questions, contact amuse@tue.mpg.de or chhatre@kth.se.
extra_gated_button_content: Request access
extra_gated_prompt: By requesting access you agree to use these artifacts only under the AMUSE non-commercial research license.
extra_gated_fields:
  Affiliation: text
  Intended use: text
  I agree to the AMUSE license terms: checkbox
tags:
  - speech-driven-gesture-generation
  - 3d-motion
  - emotion
  - smplx
---

# AMUSE-BEAT

Paper page: [Hugging Face Papers 2312.04466](https://huggingface.co/papers/2312.04466)  
ArXiv: [2312.04466](https://arxiv.org/abs/2312.04466)

This gated dataset repo hosts the released AMUSE-BEAT processed artifacts used for AMUSE.

## What is included

- `BEAT-processed/eng_data_processed/`
- `BEAT-processed/processed-all-modalities/aligned-dtw/`
- `BEAT-processed/processed-all-modalities/fbanks/disentagler_loader.npz`
- `beat_annotations_english/beat_cut_sem/`
- `lmdb/data.mdb`
- `train.csv`: a lightweight manifest loadable with `datasets`

## Quick start

```python
from datasets import load_dataset

ds = load_dataset("kiranchhatre/amuse-beat", split="train")
print(ds[0])
```

The root `train.csv` is a manifest for discovery and inspection. The full processed artifacts are
also stored in this repo and can be downloaded with `snapshot_download(...)` after access is granted.

## Redistribution notes

- Third-party assets such as SMPL-X and Blender resources are not redistributed here.
- This repo contains AMUSE-BEAT processed artifacts and annotations only.

## Scope clarification

- If you want to use the released prebuilt AMUSE training artifacts, this repo contains the processed data plus the LMDB path used by AMUSE.
- If you want to reconstruct everything from original raw BEAT and SMPL-X-style sources, this HF dataset repo is not the full raw-data release; follow the full setup and data instructions in the AMUSE GitHub repository: https://github.com/kiranchhatre/amuse

## Citation

If you use these artifacts, please cite AMUSE:

If you use AMUSE-BEAT as well, please also cite the original BEAT dataset and the EMAGE project as noted in the AMUSE release materials.

```bibtex
@InProceedings{Chhatre_2024_CVPR,
    author    = {Chhatre, Kiran and Daněček, Radek and Athanasiou, Nikos and Becherini, Giorgio and Peters, Christopher and Black, Michael J. and Bolkart, Timo},
    title     = {{AMUSE}: Emotional Speech-driven {3D} Body Animation via Disentangled Latent Diffusion},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {1942-1953},
    url       = {https://amuse.is.tue.mpg.de},
}
```